Fast hierarchical inversion for borehole resistivity measurements in high-angle and horizontal wells using ADNN-AMLM

Abstract

With the rapid development of deep learning, intelligent schemes have been gradually introduced to solve various inverse nonlinear problems. In this paper, we combine an efficient adaptive deep neural network (ADNN) framework with an adaptive modified Levenberg-Marquardt (AMLM) algorithm based on a three-layer inversion model to exact the formation resistivity and invasion depth from the measurements of array laterolog. The ADNN presented in this paper can achieve the 2D/3D fast forward modeling of array laterolog. The AMLM algorithm and a hierarchical inversion scheme are adopted to improve the anti-noise ability and convergence in complex logging environments, as well as achieve the fast and accurate reconstruction of longitudinal resistivity profiles in high-angle (HA)/horizontal (HZ) wells. Numerical simulations show that ADNN-based forward modeling only takes 0.021 s for each logging point, and the maximum relative error is less than 2%. The three-layer inversion model can eliminate the effect of the surrounding bed and improve the inversion accuracy in thinly layered formations. The AMLM inversion algorithm can effectively suppress the influence of noise, and takes only 10 steps to achieve convergence.